Perplexity (NLP)

In Depth

In natural language processing, perplexity quantifies how well a probability model predicts a sample of text. AI-generated text tends to have low perplexity because it follows the statistical patterns the model learned during training. Human text, with its creative choices, errors, and unpredictable phrasing, typically shows higher perplexity. AI detectors use perplexity as one of several signals to classify text origin.

Examples

• A GPT-4 output with perplexity of 15 (low, likely AI)
• A human blog post with perplexity of 85 (high, likely human)
• Edited AI text with perplexity of 45 (ambiguous zone)

Try It Yourself

Related Terms

Frequently Asked Questions